166
X. Jin
(SAS Institute Inc. 2016) to conduct a series of topic modeling, semantic analysis, and word-cloud analyses. JMP Pro13 (SAS Institute Inc. 2016) uses a bag of
words approach to conduct topic analysis in which the analysis is built on the word
and phrase calculation. Specifically, topic analysis is performed by conducting a
varimax rotated singular value decomposition of the document term matrix (DTM,
see Klimberg and McCullough 2016).
Before generating the DTM and conducting a topic analysis, Regex tokenizing was
applied to transform text to lowercase, identify the regular expressions, and remove
the most punctuations. Furthermore, a stop-word list was created and applied to clean
data. A Latent Semantic Analysis (LSA) was used to extract and represent terms and
phrases from a corpus so that the DTM was reduced to a manageable size for further
analysis. A Singular Value Decomposition (SVD) was generated to decomposes the
DTM into three matrices:
DTM = D * S * T.
These matrices are defined as follows:
• D is an orthogonal document-document matrix of eigenvectors.
• T is an orthogonal term-term matrix of eigenvectors.
• S is a diagonal matrix of singular values (Klimberg and McCullough 2016, p. 338).
While generating the curated DTM, natural language processing techniques were
used. Later, text clustering, principal components, and factor analysis were utilized
to interpret the DTM (Klimberg and McCullough 2016). Lastly, topic analysis and
word-cloud were generated. Topic analysis can calculate how much each tweet
contributes to each topic. It can also discover the most popular topics of a specific
event based on word frequency calculations. With this approach, this case study was
able to identify the dominant topics regarding Hurricane Maria in its resolution stage.
9.5 Results
To capture the major topics associated with Hurricane Maria, this study found the top
hashtags that were frequently used in the resolution stage (November 6–November
21, 2017). The top three hashtags were #puertorico (459), #theoriginals (381), and
#hurricane (151). The top three URLs used in the tweets were identified as well. The
first link was a denotation webpage created by the disaster relief and recovery program
of UNIDOS (2017): https://hispanicfederation.org/unidos. The second link was about
an article calling for volunteers for Puerto Rico, which is written by Navarro (2017):
https://trib.al/6FI4dfx. The third link was an article (Dickerson 2017) focusing on the
mental health crisis of Puerto Rico community after Hurricane Maria (https://www.
nytimes.com/2017/11/13/us/puerto-rico-hurricane-maria-mental-health.html).
To further understand the communication patterns that emerged from the resolution stage of Hurricane Maria, this study combined topic modeling, semantic analysis content analysis, and word-cloud. To answer the research question, five topics
X. Jin
(SAS Institute Inc. 2016) to conduct a series of topic modeling, semantic analysis, and word-cloud analyses. JMP Pro13 (SAS Institute Inc. 2016) uses a bag of
words approach to conduct topic analysis in which the analysis is built on the word
and phrase calculation. Specifically, topic analysis is performed by conducting a
varimax rotated singular value decomposition of the document term matrix (DTM,
see Klimberg and McCullough 2016).
Before generating the DTM and conducting a topic analysis, Regex tokenizing was
applied to transform text to lowercase, identify the regular expressions, and remove
the most punctuations. Furthermore, a stop-word list was created and applied to clean
data. A Latent Semantic Analysis (LSA) was used to extract and represent terms and
phrases from a corpus so that the DTM was reduced to a manageable size for further
analysis. A Singular Value Decomposition (SVD) was generated to decomposes the
DTM into three matrices:
DTM = D * S * T.
These matrices are defined as follows:
• D is an orthogonal document-document matrix of eigenvectors.
• T is an orthogonal term-term matrix of eigenvectors.
• S is a diagonal matrix of singular values (Klimberg and McCullough 2016, p. 338).
While generating the curated DTM, natural language processing techniques were
used. Later, text clustering, principal components, and factor analysis were utilized
to interpret the DTM (Klimberg and McCullough 2016). Lastly, topic analysis and
word-cloud were generated. Topic analysis can calculate how much each tweet
contributes to each topic. It can also discover the most popular topics of a specific
event based on word frequency calculations. With this approach, this case study was
able to identify the dominant topics regarding Hurricane Maria in its resolution stage.
9.5 Results
To capture the major topics associated with Hurricane Maria, this study found the top
hashtags that were frequently used in the resolution stage (November 6–November
21, 2017). The top three hashtags were #puertorico (459), #theoriginals (381), and
#hurricane (151). The top three URLs used in the tweets were identified as well. The
first link was a denotation webpage created by the disaster relief and recovery program
of UNIDOS (2017): https://hispanicfederation.org/unidos. The second link was about
an article calling for volunteers for Puerto Rico, which is written by Navarro (2017):
https://trib.al/6FI4dfx. The third link was an article (Dickerson 2017) focusing on the
mental health crisis of Puerto Rico community after Hurricane Maria (https://www.
nytimes.com/2017/11/13/us/puerto-rico-hurricane-maria-mental-health.html).
To further understand the communication patterns that emerged from the resolution stage of Hurricane Maria, this study combined topic modeling, semantic analysis content analysis, and word-cloud. To answer the research question, five topics